Due to the severe shortage of data in the case records of hazardous materials road transport accidents, a de-completion model with missing data based on attribute distribution probability was developed to complete the missing data. Adopting Gradient Boosting Decision Tree (GBDT) as the base learner, the Bagging algorithm constructs an integrated GBDT prediction model. Besides, the completed data is used to the accident type prediction, and the analysis of factors influencing the accuracy is based on the predicted outcomes. Compared with other completion methods, the results reveal that the mean square error (MSE) value of the data after the completion of the attribute distribution probabilistic completeness model is the smallest. Applying the integrated GBDT prediction model, the accident type prediction of accuracy rate can reach 87.3%. Seven accident characteristic values have a notable influence on the accuracy of accident type prediction. According to various accident types, the values of the incident eigenvalues that determine the prediction accuracy are distinctive. By contrast, the characteristic value of "the number of vehicles involved" has a notable influence on the accuracy of most accident type predictions.

CICTP